preprint - unisa
TRANSCRIPT
Preprin
t
Enhancing Computer Students’ Academic Performance
through Explanatory Modeling
Full Paper
SACLA 2019
© The authors/SACLA
Leah Mutanu1[0000-0001-5164-6424] and Philip Machoka2[0000-0001-7008-7111]
1,2 United States International University - Africa, 14634-00800, Nairobi, Kenya
{lmutanu1,pmachoka2}@usiu.ac.ke
Abstract. A key challenge facing today universities is the growing attrition rate
of computer studies students, attributed to poor academic performance. While
extensive research has been conducted on how to enhance student’s performance
in computer programming, very little research investigates other computer
courses, more so in sub-Saharan Africa. This study set out to address this gap by
conducting experiments that revealed some of the factors that influence a stu-
dent’s overall academic performance at university through explanatory modeling.
Results obtained showed that a student’s background in mathematics and their
performance in the Introduction to Information Systems course were key in de-
termining performance. Unexpectedly, prior computer skills or secondary school
grades had less impact. The strategies identified for enhancing student’ perfor-
mance include an emphasis on building a student’s mathematics background,
providing a string teaching approach to foundational computing courses, re-struc-
turing of courses in the computer program, and linking courses across the curric-
ulum. Therefore, explanatory modeling creates an opportunity to adopt a proac-
tive approach to enhancing the performance of computer studies students.
Keywords: Computer Science Education, Academic Performance, Explanatory
Modeling.
1 Introduction
Globally, Higher Education is the fastest growing segment of post-secondary education.
However, this sector faces a myriad of challenges. Key among them is student retention.
The challenges, however, vary from region to region. In the USA, for example, the
Preprin
t
2
challenge is retaining more women and people from underrepresented minorities (Af-
rican-Americans, Hispanics, and Native Americans) in computer-related studies. In
Kenya, education quality is cited as one of the key challenges facing public university
education [1]. Universities also face low enrolment numbers in computer-related degree
programs as compared to other degree programs. Another challenge is the high number
of student dropout rates. While various reasons exist why students drop out, student
academic performance has been identified as one of the biggest drivers [2]. A study by
Njoroge et al. [3] investigating student attrition rates in private universities in Kenya
revealed that academic performance contributed to increased attrition rates. The study
recommended mechanisms to be put in place for early detection of attrition risk sup-
ported by technology to ensure students pursue their studies to completion.
The objective of this research was to pursue Njoroge et al. recommendation, that is,
to identify factors in a student’s learning environment that can serve as indicators of
students’ academic performance. This research focused on academic factors that im-
pede students' performance in computer-related studies. The remainder of the paper is
organized as follows: Section 2 outlines a review of related work; Section 3 discusses
the research methodology; Section 4 presents the findings and discussions; Section 5
highlights the key limitations of the study, and Section 6 concludes and makes recom-
mendations for future work.
2 Review of Related Work
High dropout rates are common in Computer related degree programs at universities.
Two main causes of the problem as pointed out by Gárcia-Mateos [4] are motivation
and complexity of these courses. In determining these problems, however, their study
did not factor in the student’s prior education background before joining a university.
The assumption was that any student who has enrolled in a computer-related program
was qualified to undertake the degree course. However, this study revealed that there
are prior academic background factors that can influence a student’s academic perfor-
mance. This study did not consider students’ social, economic, cultural, and geographic
factors because their solution requires a different approach. A study by Kumar [5] found
that different demographic groups (economic status, gender, race, major, and type of
institution) required different intervention approaches in Computer Science. This
study, therefore, set out to explore a students’ prior academic background before join-
ing university rather than their demographic groups.
A study conducted by AlMurtadha et al., [6] in Saudi Arabia investigated the key
factors that influenced computer studies students’ academic performance. Computer
studies students are those that are taking a degree program in any computer related
studies such as Computer Science and Information Systems. A number of factors were
identified in the study. Some of the factors included: age, gender, student major, means
of transport to school, parents education level, English proficiency level, sitting posi-
tion, fear of exam, study-schedule, drug abuse, daily-sleeping hours, twitter account
use, sports engagement, hobbies, and community service engagements. The study re-
vealed that a student’s major influenced their academic performance. Other factors that
Preprin
t
3
influenced performance included their age and gender. A student’s academic back-
ground in English language and parents’ education level partly influenced academic
performance while their campus social life did not. From this study, it can be observed
that a number of factors influence students’ academic performance. These factors can
be categorized into social, economic, and academic. While all those factors are signifi-
cant, this study focuses only on those factors that academic institutions can influence,
i.e. the academic factors, through the provision of remedial courses or alternative teach-
ing approaches. They include the students’ academic background in Mathematics,
Computer and English courses. A research study by Nash [7] points out that Students
joining Tertiary education are not always equipped with the skills that can enhance their
learning experience. University students are required to use computers to access course
materials, write assignments with good grammar, and perform calculations. To be able
to cope with these demands, students need a general understanding of computer, lan-
guage and mathematics skills. Additionally, factored in this study is the students’ aca-
demic performance at university.
A study done by Garcia and Al-Safadi [8] on factors affecting students' academic
performance in computer programming found that classroom management skills were
key to improving student performance. The study, however, focused only on the per-
formance of one subject – computer programming and not the entire degree program.
More recently, Wang et al., [9] also conducted a similar study that focused on enhanc-
ing students’ computer programming performance. They recommend a more compre-
hensive study that focused on other courses offered in the degree program rather than
only on one subject. While there is a lot of research on how technology enhances stu-
dents’ performance, very little research focusses on how to enhance student perfor-
mance in computer-related studies. To the best of the authors’ knowledge, most of the
existing research in the area identified factors influencing only a single subject that is
computer programming, ignoring other subjects in the degree programs. This study ad-
dresses this gap by looking at performance in all courses in the entire degree program.
Furthermore, there are limited studies in the area focusing on institutions within sub-
Saharan Africa.
The work of Garcia and Al-Safadi [8] that investigated factors affecting students'
academic performance was based on students’ perceptions. They concluded that in-
structors’ classroom management skills such as preparation for the topic and teaching
techniques influenced performance. Instructional materials did not have a big influence.
A related study that also sought student’s reflection revealed that education pedagogy
was a key factor in influencing students’ performance in computer related courses [10].
In their work, Barlow-Jones and Westhuizen [11], show the relationship between uni-
versity pre-entry attributes and student performance in computer programming. Their
findings revealed that there was a correlation between previous programming experi-
ence and performance in programming modules while there was no correlation between
the socio-economic status, educational background, high school Mathematics, and Eng-
lish scores. Their study however also collected data from students’ opinions rather than
examining actual students’ scores. This research presents an alternative approach that
does not focus on students’ perceptions but on student’s actual performance. In effect,
it paints a more accurate picture of the background factors that influence performance.
Preprin
t
4
Computer students often require a special set of digital skills that other degree pro-
grams do not demand. This is especially challenging in developing countries where the
digital divide gap is large. The lack of infrastructure and low household income often
denies students the opportunity to engage with technology adequately especially during
their formative years of schooling. Chikumba [12] highlights the extent of the problem
in Malawi where private secondary schools performed better in computer studies than
public secondary schools due to poor investment in computers, teaching materials and
staff required to deliver the subject. Due to this challenge, it is not a requirement for
students joining many universities in developing countries offering computer-related
degree programs to have the requisite technical background. However, some courses
can be recommended for students intending to take computer related studies. For ex-
ample, Akinola and Nosiru, [13], recommends a priori knowledge of Physics and
Mathematics is essential for a student to excel better in Computer related studies. They
also mention that better teaching methods and techniques can enhance student perfor-
mance by changing students’ perception of computer-related courses. A different study
on factors promoting success in Computer Science by Wilson [14] revealed the Com-
puter Science performance predictive factors as the student’s comfort level and math
background. No significant influence was based on gender differences. The perfor-
mance in the introduction to programming courses also had a positive influence on suc-
cess. Research on how language influences performance revealed that contrary to the
generally accepted view that achievement in high school mathematics courses is the
best individual predictor of success in undergraduate Computer Science, success in
English correlates better with the actual performance [15]. Their study, however, was
conducted in a social context where many students were not native English speakers.
In their study, Akinola and Nosiru, [13] used Fuzzy sets operation approach to arrive
at the factors that influence performance. They posit that selecting the factors that in-
fluence students' performance in computer programming is not an easy task as it in-
volves human decision making which can be imprecise or subjective. This problem
cannot be handled effectively by probability theories. Fuzzy set handles problems that
involve the need to seek consensus among many decision makers. Students gave their
opinions according to their own criteria for each factor by selecting a value and the
union of their evaluations to all the currently available alternatives was represented in
the form of a fuzzy set. They state that the approach helped in eliminating outlier deci-
sion-making, hence, presenting a more accurate and reliable result. Although faculty
(instructors) have traditionally found ways of identifying performance challenges, there
is a need to enhance the process through innovative ways of modeling data. The objec-
tive of the research was to model student data, captured over a period, with a view of
providing causal explanations to performance for purposes of early intervention. Data
mining techniques have been used extensively in research to develop early warning
systems for risk aversion. Within the health domain, early warning systems have been
defined as surveillance systems that collect information on epidemic-prone issues such
as diseases in order to trigger prompt public interventions [16]. Fuzzy logic is used to
map risk patterns. The research described in this paper therefore aimed at modeling
computer students' academic performance in order to find ways of enhancing it.
Preprin
t
5
3 Methodology
The study was conducted at a private university in Kenya. The university offers two
computer-related degree programs, Computer Science and an Information Systems.
Both programs had the same pre-entry admission requirements for secondary school
final exam grade point average (GPA), however, the Computer Science program re-
quired higher secondary school scores in Mathematics and Physics for admission into
the program. Secondary data was collected on students taking both programs for a pe-
riod of five years. Only data recorded in the Students’ Information Systems (SIS) could
be used i.e. the student’s GPA at secondary school, placement tests score (skills assess-
ment score on admission to university), student major, gender, current year of study,
year of admission, current GPA for courses taken so far at the university, and the coun-
try of origin. The research focused on academic factors that the institutions could influ-
ence or use to influence practices at the university to enhance the students’ academic
performance. This research aimed at modeling computer students' academic perfor-
mance in order to find ways of enhancing it. It achieved this through two research
objectives; (i) Identification of academic factors influencing student academic perfor-
mance in computer-related studies, and (ii) Establishing ways of enhancing the com-
puter-related degree programs curriculum to address academic factors influencing stu-
dent academic performance in computer-related studies.
The data set constituted 6000 students who had joined the university over a five-year
period (2014 – 2018). A random sample was required for analysis. The study used the
formula in equation 1 was used to calculate the study sample size.
𝑆𝑎𝑚𝑝𝑙𝑒𝑆𝑖𝑧𝑒 =
𝑧2 x 𝑝 ( 1 − 𝑝)𝑒2
1 + ( 𝑧2x 𝑝(1 − 𝑝)
𝑒2𝑁)
(1)
N = population size • e = Margin of error (percentage in decimal form) • z = z-score • p = sample proportion
A confidence level of 95% was used which requires a Z-score of 1.96. The z-score is
the number of standard deviations a given proportion is away from the mean. A 3%
margin of error was selected. The sample size was therefore obtained as 907. The study,
therefore, picked 1000 student from a population of 6000 students’ records to work
with. This sample was picked to act as a test set for experiments in predictive modeling
however, future experiments will focus on using the entire sample size. However,
through the data cleaning process, several records were discarded giving a final sample
size of 858 students. This number was found to be acceptable as it still gave the desired
3% margin of error.
For ethical purposes, identifying features such as the students’ ID number, students
name and addressed were filtered from the data. An SQL script was written and used
for the extraction to ensure that all the data was extracted anonymously. The resultant
data was placed on an excel spreadsheet for data cleaning purposes and subsequent
analysis. To ensure that the data was an accurate representation of the five years, the
year of admission was also retrieved. Cleaning of the data was done using an MS-Excel
Preprin
t
6
Spreadsheet enable records with data entry errors or missing data to be dropped. From
the data cleaning exercise, 142 records were dropped. Data Analytics was then con-
ducted with open source data mining tools Weka [17] and R [18]. The IBM SPSS
statistical analysis tool [19] was also used to explore and validate the results. The use
of several tools enriched the process by validating the results obtained. Clustering and
Decision tree algorithms were used to classify the data. From the Data mining exercise,
patterns in the data were identified and used for explanatory modeling. The research
design was descriptive where the characteristics of correlations between two or more
entities were explored and visualization techniques used to represent the data. Quanti-
tative research techniques were used to emphasize objective measurements and the sta-
tistical, mathematical, or numerical analysis of data collected through surveys, or by
manipulating pre-existing statistical data using computational techniques. In this re-
search data was gathered and used to generalize across groups of students to explain
performance.
4 Findings and Discussions
This section identifies patterns in the data for purposes of modeling the performance of
students taking undergraduate computer-related degree programs at the university. It
describes the techniques used and the results obtained.
4.1 Factors Influencing Cumulative GPA
The research set out to identify some academic factors that influence students’ perfor-
mance based on existing literature described in section 2. Specifically, the research an-
alyzed factors such as the student’s final GPA at secondary school, placement tests
score on admission to the university, students major, gender, current year of study, year
of admission, test scores for courses done at the university, and the country of origin.
Placement tests were given to students upon joining the university. Scores from these
tests are used to assess student's prior knowledge in Mathematics, English, and compu-
ting skills upon admission. Students who do not meet the cut off points are required to
take remedial classes for a semester in the respective courses. The prior mathematics
knowledge that computer programs require include basic algebra and statistics. The
degree programs also offer additional mathematics courses such as discrete mathemat-
ics and algebra. Most of the computer courses offered at the university have Mathe-
matics courses as a pre-requisite. These include courses such as Data Structures, Deci-
sion Analysis, and Data analytics. However, the first computer programming course – Fundamentals to Programming Logic - assumes no prior knowledge of programming
and does not have any mathematics prerequisites. The focus of the course is on impart-
ing procedural programming skills.
Distribution of the frequencies for the data used in this study is represented in Fig.
1. The figure shows that there were more male students in both programs. Given a pass
mark of 60% majority of the students managed to pass the placement tests in IST (Com-
Preprin
t
7
puter placement) and English. The performance in Mathematics was not as good, how-
ever, it should be noted that Computer Science competes with other Science and Engi-
neering disciplines for students with a good Mathematics Background.
(a) Computer Science Program
(b) Information Systems Program
Fig. 1. Sample Frequency Distributions of the Datasets used for the study.
(a) EM Clustering Algorithms (b) K-Mean Clustering Algorithm
The data was analyzed using clustering algorithms to give an indication of which fac-
tors influenced the cumulative GPA. For this research Expectation–Maximization (EM)
clustering technique [20] was preferred because it provides better optimization than
distance-based or hard membership algorithms, such as K-Means [21]. EM easily ac-
commodates categorical and continuous data fields, thus, making it the most effective
technique available for proper probabilistic clustering. K-Means clustering is a method
Fig. 2. Sample Frequency Distributions of the Datasets used for the study.
Preprin
t
8
of vector quantization that partitions observations into clusters based on the nearest
mean, serving as a prototype of the cluster. K-Means was also used for purposes of
validating the results. The Data had to be converted into numeric representations to
perform K-mean clustering.
Determining the Significance of Factors Used. Cluster analysis showed that some
variables have a bigger impact on the student’s cumulative GPA than others. For ex-
ample, Fig. 2 shows that a higher math placement score resulted in a higher student’s
cumulative GPA. Both EM and K-Means clustering algorithms gave the same results.
Similar results were also observed for the student’s year of study where students at their
third and fourth years of study had higher cumulative GPAs than those in their first and
second years of study.
However, similar tests showed that the English and Information Systems Technol-
ogy (IST) placement scores had less impact on the student's cumulative GPA as de-
picted in Fig. 3. This showed that students do not require prior knowledge in Computers
to perform well in computer-related degree programs. Prior knowledge in computers
refers to computer literacy skills. The cluster centers of high cumulative GPA scores
were close to the 50% test scores. However, further investigation to understand why
factors such as previous computer knowledge were not influential contrary to popular
belief is required. Similar results were also observed for the student's secondary school
final exam GPA despite the expectation that students who performed well in high
school would also perform well at university. The secondary school final GPA's did not
have a big impact on the cumulative GPA scores at university. This is expected given
that the students selected for the program had similar high school final GPA scores
based on the pre-entry requirements.
(a) IST Placement Test (b) English Placement Test
Other factors examined such as the student’s gender, degree program (Computer Sci-
ence or Information Systems), and country of origin had almost no visible impact at all
when a cluster analysis was run. These findings concur with the findings of a similar
study on factors promoting success in Computer Science by Wilson [14]. However,
Fig. 3. Impact of IST and English placement scores.
Preprin
t
9
further investigation to understand why factors such as gender were not influential con-
trary to popular belief is required. The Independent Samples t-test [22] was used to
compare the difference in cumulative GPA means where two independent groups ex-
isted such as male and female students or Computer and Information Systems degree.
The purpose of the t-test was to determine whether the difference between the two
groups was statistically significant. A t-test conducted on the student’s cumulative GPA
versus student’s gender showed that the difference in cumulative GPA was not statisti-
cally significant between the two groups as tabulated in Table 1 (a) The results
t(323.035) = -7.300, p = 0.000, show that there is no statistically significant difference
in the variances between the two groups. The p-value is below the critical significance
level of 0.05. Running a similar t-test on the student's degree program versus the cumu-
lative GPA also confirmed that there was no statistically significant difference in the
GPA reported between the two groups (Table 1 (b)). The results in the table, t(664.239)
= 5.017, p = 0.000, show that there is no statistically significant difference in the vari-
ances between the two groups. The p-value is below the critical significance level of
0.05. The results statistically confirmed that gender and the degree program had no
significant influence on the student’s cumulative GPA.
Table 1. Determining the significance of gender and degree program.
(a) Student Gender
(b) Student Degree
It is worth noting that further analysis showed that there were more male students ad-
mitted in the two programs although female students had comparatively higher cumu-
lative GPA’s as illustrated in Table 2 (a). The difference, however, was not statistically
significant. Both computer-related degree programs had low female enrolments. Addi-
tionally, Students in the Information Systems Degree program were more than those in
the Computer Science program. This was expected because the former program had
been offered for a longer period. However, it was observed that the Computer Science
program had students with slightly higher cumulative GPA than those from the Infor-
mation Systems program as illustrated in Table 2 (b) although the difference was also
not statistically significant.
Preprin
t
10
Table 2. Impact of gender and degree program on cumulative GPA. Male Female
Mean Cumulative GPA 2.46533 2.892108
Total 673 185
(a) Impact of gender and degree program on cumulative GPA.
Information
Systems
Computer
Science
Mean Cumulative GPA 2.659755 2.387734
Total 535 323
(b) Impact of gender and degree program on cumulative GPA.
Table 3. Statistical significance using Pearson Correlation.
(a) Correlation (Math placement test and cumulative GPA
cum_gpa Math
placement
cu
m_
gp
a Pearson
Correlation 1 .364**
Sig. (2-tailed) .000
N 857 857
Math
pla
cem
ent Pearson
Correlation .364** 1
Sig. (2-tailed) .000
N 857 857
**. Correlation is significant at the 0.01 level (2-tailed).
Preprin
t
11
(b) Correlation (IST placement test and cumulative GPA
To assess the significance of each of these factors a Pearson Correlation [23] analysis
was conducted. The bivariate Pearson Correlation produces a sample correlation coef-
ficient, r, which measures the strength and direction of linear relationships between
pairs of continuous variables. The Pearson Correlation was used to present the statisti-
cal evidence for variables that impact the cumulative GPA or not. The results of the
correlation confirmed the observations made from the clustering tests that mathematics
placement tests scores influenced the student’s cumulative GPA. There is a statistically
significant correlation between Mathematics placement test scores and a student's cu-
mulative GPA. The Pearson’s r is positive indicating that when one variable increases
the second variable also increases. The Sig (2-Tailed) value is less than or equal to 0.05
showing that the relationship is statistically significant. However, the correlation be-
tween a student’s IST placement test score and a student’s cumulative GPA was not
statistically significant (Table 3) because the value of the Pearson’s coefficient r is close
to zero. The correlation between a student’s country of origin and a student’s cumula-
tive GPA was not statistically significant.
Mining Patterns in the data. The results obtained so far identified variables that were
significant in determining a student’s cumulative GPA and the extent to which each
variable impacted the cumulative GPA. They, however, do not describe patterns in the
data showing how all the variables jointly impacted the cumulative GPA. This is useful
when modeling relationships between variables. In order to describe patterns in the
data, further analysis of all the variables was required to show how they influence each
other. For this task, Clustering and decision tree algorithms were applied to the data.
When a cluster analysis was conducted with the EM clustering algorithm, three large
clusters were identified. The results presented in Fig. 4 show three significant clusters
labelled 2 (GPA-2.6), 6 (GPA-2.5), and 8 (GPA-2.1) as having the highest cluster den-
sities i.e. 22%, 25% and 22% respectively. These clusters are groups of students with
similar characteristics. The math placement scores for each cluster were observed to be
59% 53% and 70% respectively (Fig. 4). The high densities showed that majority of the
students had these characteristics. The figure further shows two clusters labeled 1 and
3 with high cumulative GPA scores of 2.9 and 3.1 respectively. The Mathematics place-
ment scores are also higher at 71% and 68% respectively. This confirms the earlier
results that math placement score influenced the cumulative GPA. Although clustering
showed the existence of relationships, it is difficult to tell at a glance what the relation-
ships were. Additional analysis was required to describe the factors that formed each
cluster.
Running decision tree algorithms on the data revealed the relationship between dif-
ferent variables as illustrated in Fig. 5. From the results, it can be observed that the
Introduction to Information Systems course (IST1020), the year of study (cl), and the
math placement test played the most significant role in determining a student’s cumu-
lative GPA. The Introduction to Information Systems course imparts general computer
literacy skills to students. It does not include any computer programming. A few second
level courses were also found to have an impact, that is Computer Organization (taken
by the Information Systems degree students) and Computer Networks (taken by the
Preprin
t
12
Computer Science degree students). The Computer Organization course exposed stu-
dents to computer architecture and assembly language while the Computer Networks
course introduced students to data communication protocols and devices. Both courses
were taken by students as soon as they completed the introduction to Information Sys-
tems course. The higher-level courses had no impact. The findings brought out the im-
portance of laying a strong foundation for students during their initial courses in com-
puting. The approach to the foundation classes served to either propel students to suc-
cess or failure. It was observed that seasoned instructors who were Professors in the
department taught higher-level courses or graduate programs while early career instruc-
tors and part-time instructors were left to handle introductory level courses in the de-
partment. This approach needed to change to enhance students’ performance.
Fig. 4. Mining patterns through clustering.
Preprin
t
13
Key: IST2010 – Computer Organization, APT2050 – Computer Networks, IST1020 – Introduction to Infor-
mation Systems
Fig. 5. Classification of factors influencing cumulative GPA using decision trees.
Table 4. Relationship between courses in a degree program.
Preprin
t
14
The study also sought to establish how courses taught at various levels influenced
each other. To achieve this a Pearson Correlation was performed across the various
courses offered. The courses are mapped against each other and the Pearson Correlation
values provided. The lower-level courses are courses offered in year 1 and year 2 indi-
cated in green in the column and row headings. The higher-level courses are courses
offered in year 3 and 4 indicated in blue in the column and row headings. The results
presented by the correlation matrix in Table 4 (a) showed that for the Computer Science
program lower level courses appeared to influence each other positively (Pearson Cor-
relation value above 0.4), such that a high grade in one would most likely mean a high
grade in the other. A closer examination revealed that this occurred between related
courses such as Computer Organizations and Operating Systems or Web Design and
Computer Programming. Related courses in this context are considered as courses that
are a pre-requisite of the other. For example, Computer Organizations is a pre-requisite
course for Operating systems and Computer Programming is a pre-requisite for Web
Design. Unexpected relationships identified were between Computer Organization and
Computer Programming which are courses that are not indicated as prerequisites of
each other. This, however, might be attributed to the fact that Assembly Language Pro-
gramming was part of the Computer Organization course.
For the Information Systems program, fewer courses seemed to influence each other
as depicted by the presence of more white cells in Table 4 (b) (Pearson Correlation
value less than 0.2). This was a possible indication of the lack of proper linking when
teaching courses across the curriculum. Some significant relationships, however, ex-
isted among related higher-level courses such as System Analysis and Design, Object
Oriented Programming, and Database Management Systems or Decision Analysis,
Data Structures and Algorithms, and Mobile Programming. The relationships were
expected because the courses were related and, therefore had been stated as pre-requi-
sites of each other in the curriculum. One unexpected relationship, however, was iden-
tified between the Digital Laboratory and Object-Oriented Programming where stu-
dents who did well in Digital Laboratory also did well in Object-Oriented Programming
and vice versa. The two courses had distinct content and, therefore, had not been indi-
cated as pre-requisites of each other in the curriculum. Further research is required to
investigate the cause of this relationship.
4.2 Strategy for Enhancing Performance
The experiments described in Section 4.1 provided insights into weaknesses in the cur-
rent degree programs offered at the university. Results from this study can offer an
opportunity to review the programs and align courses properly to enhance student’s
academic performance in computer-related programs. Strategies on the courses to be
considered when admitting students into the program, the courses to be emphasized or
restructured in the curriculum, and ways of reducing attrition rates through performance
are discussed in this section.
Preference on Student Mathematics Background. From the analysis conducted,
the findings showed that the results of the Mathematics placement test scores had a
Preprin
t
15
positive impact on the student’s cumulative GPA. Students with a strong Mathematical
background performed better than students who were weaker in the subject. This was,
however, not the case with the computing and English skills backgrounds. Courses in
computer-related studies, therefore, need to admit students with a strong mathematical
background. As mentioned earlier, Computer Science and Information Systems com-petes with Science and Engineering disciplines for students with a good Mathe-matics background. Therefore, an alternative strategy to attract students with ex-cellent Mathematics background needs to be sought. Additionally, there is a need to
add mathematics courses such as Discrete Mathematics, Formal Logic, Differential,
and Integral Calculus to make up for weaknesses in the required areas.
Emphasis on Introduction to computers course. Another key area that influenced
the student’s cumulative GPA is their performance in the Introduction to Information
Systems course. Students who performed well in this course ended up with good overall
cumulative GPA scores. Emphasis, therefore, needs to be put on how the courses are
taught to develop student’s interest as well as lay a good solid foundation for the two
computing programs. Basic skills that will be required in advanced courses should be
instilled at the foundation level. Further research on the specific skills required at this
level is required. Such skills should, therefore, be integrated into the Introduction to
Information Systems course. Further senior and experienced instructors who tradition-
ally teach higher level courses and graduate courses should be encouraged to take
lower-level courses to ground and mentor the students well during foundation courses.
Structure of the curriculum. The results further showed that there was a significant
positive correlation among courses in the Computer Science degree. Students who did
well in the introductory programming courses also did well in the advanced program-
ming course. This was, however, not the case for the Information Systems degree pro-
gram. This was attributed to the fact that the Computer Science degree was developed
after the Information Systems degree and, therefore structured better to accommodate
lessons learned from the Information Systems degree. The structure of the Information
Systems degree program has since been revised to address some of these challenges. It
is envisaged that the results of these structural changes will reflect in future studies.
Linking Courses across the curriculum. The researchers also noted the need to
relate courses when teaching the program as well as re-examining prerequisites in the
program. While relationships were observed among programming courses, the same
was not apparent among networking courses or computer security courses. Students
who did well in the lower-level networking and security courses should also be able to
do well in similar courses at the advanced level. The absence of well-defined relation-
ships could, however, be due to the lack of courses at a higher level to advance courses
in those tracks. Further research is required to establish the reason for the lack of these
relationships.
Prediction modeling for an early warning system. Machine Learning and Data
Mining techniques can be combined to identify weaknesses in a degree program that
may not be easy to detect manually. Through prediction modeling, forecasts on stu-
dents’ performance can enable stakeholders to proactively take measures to enhance
student’s performance. Automation of the process also eliminates human biases that
can interfere with the process. It presents an opportunity for instructors to identify
Preprin
t
16
weaknesses in the programs offered with minimal effort. Through the provision of early
warnings, student attrition rates can be reduced and subsequently the quality of educa-
tion can be improved.
The paper performed explanatory modeling to reveal some of the factors that influ-
ence academic performance. The suggested strategies are based on anecdotal evidence
about the actual effectiveness and efficiency of the suggested solution. They, however,
form a basis for further research to investigate the effectiveness of the suggested ap-
proaches.
5 Limitations of the Study
This study investigated the academic performance of students taking computer-related
degree programs at the university. The general objective of the study was finding out
whether factors influencing student academic performance can be modeled to provide
explanations on students’ performance with a view of enhancing their performance.
Although the study was conducted in a single private university, the results have the
potential to be generalized for other institutions. Further research in this area should
include a comparative study among several institutions to provide recommendations
for enhancing the approach.
Traditionally many universities emphasize that mathematics and physics skills are
required for admission into computer-related degree programs. Findings from this re-
search confirmed that background skills in mathematics were a key factor in determin-
ing a student’s performance. The results provide a basis for investigating students’ per-
formance. Further research is required to describe the correlations with a view of estab-
lishing whether they are pointers to actual correlations or simply mere casualties. Fur-
ther, the study did not investigate the relationship between the student’s background
knowledge in physics and their academic performance. This is because physics skills
were not assessed by the university and no records of the students’ secondary scores
for physics were captured in the university’s Student Information System.
Student's academic performance is influenced by several factors originating from
their social, economic, and academic backgrounds. This research, however, only fo-
cused on academic factors arising from a student's educational background as well as
their current academic performance in courses taken at the university. These were fac-
tors that could easily be collected from the Student’s Information Systems. Addition-
ally, they are factors that the institution can address to enhance the student’s academic
performance.
Findings from the study revealed the need to improve on the teaching approaches
for lower level courses as they have an impact on the student’s cumulative GPA later
in the program. However, the study did not investigate how to do this. Further research
is required to explore how different teaching approaches influence the cumulative GPA.
Preprin
t
17
6 Conclusions and Future Research Directions
A review of existing literature revealed that few similar studies had been done. Most of
the research focused on enhancing the performance of specific courses in computing
programs rather than all courses across the curriculum. Several studies conducted used
students’ perceptions which is important but might not always give the actual picture.
This study set out to address this gap by analyzing performance across all courses in a
program with an emphasis on factors that the institution can influence. Further, it made
use of actual student scores to evaluate factors that influence performance. Data ana-
lytics techniques were used to classify data and provide explanations. The results ob-
tained showed that a student’s mathematics background and their performance in their
introductory computing course were key in determining performance in their computer
studies. Unexpectedly, prior computer skills or secondary school grades had less influ-
ence on a student’s performance. The non-academic factors measured such as country
of origin and gender were also not significant in determining a student’s academic per-
formance. The research also suggested strategies for enhancing student performance in
computer-related degree programs.
Further research in this area should be conducted to validate the effectiveness of the
suggested strategies. More extensive research should also be conducted among both
private and public universities in the region as well as among other degree programs.
Other factors that influence a student’s academic performance such as their social, cul-
tural, and economic backgrounds would also enhance the findings of research in this
area. Investigating the influence of other academic factors such as student’s background
in physics would also enhance the study. In this study, explanatory modeling laid the
foundation for our current research in predictive modeling. The goal was to show that
explanatory modeling can be used to identify factors that influence performance hence
setting the case for predictive modeling of academic performance. The use of prediction
models such as Regression Analysis and Neural Networks can provide a proactive ap-
proach to the problem. While this research is not exhaustive it presents an opportunity
for researchers in computer studies education to find ways of enhancing student’s aca-
demic performance through explanatory and predictive modeling techniques.
References
1. Odhiambo, G. O.: Higher education quality in Kenya: A critical reflection of key challenges. Quality in
Higher Education, 17(3), 299-315 (2011).
2. Giannakos, M. N., Pappas, I. O., Jaccheri, L., & Sampson, D. G.: Understanding Student Retention in
Computer Science Education: The role of environment, gains, barriers, and usefulness. Education and
Information Technologies, 22(5), 2365-2382 (2017).
3. Njoroge MM, Wang’eri T, Gichure C.: Examination Repeats, Semester Deferments and Dropping Out
as Contributors of Attrition Rates in Private Universities in Nairobi County Kenya. International Journal
of Education and Research, 4(3) (2016).
4. Gárcia-Mateos G, Fernández-Alemán J. L.: A course on algorithms and data structures using on-line
judging. In ACM SIGCSE Bulletin, Vol. 41, No. 3, 45-49. ACM (2009).
Preprin
t
18
5. Kumar AN.: Need to consider variations within demographic groups when evaluating educational in-
terventions. In ACM SIGCSE Bulletin Vol. 41, No. 3, pp. 176-180. ACM (2009).
6. AlMurtadha YM, Alhawiti KM, Elfaki AO, Abdalla OA.: Factors Influencing Academic Achievement
of Undergraduate Computing Students. International Journal of Computer Applications, 975:8887
(2016).
7. Nash J.: Computer skills of first-year students at a South African university. In Proceedings of the 2009
Annual Conference of the Southern African Computer Lecturers' Association 2009 Jun 29, pp. 88-92.
ACM (2009).
8. Garcia, R.A., & Al-Safadi, L.A.: Comprehensive Assessment on Factors Affecting Students' Perfor-
mance in Basic Computer Programming Course towards the Improvement of Teaching Techniques.
International Journal for Infonomics. 6. 682-691. 10.20533/iji.1742.4712.2013.0079 (2013).
9. Wang, X. M., Hwang, G. J., Liang, Z. Y., & Wang, H. Y.: Enhancing students’ computer programming
performances, critical thinking awareness and attitudes towards programming: An online peer-assess-
ment attempt. Journal of Educational Technology & Society, 20(4), pp. 58-68 (2017).
10. Pretorius H. W, Hattingh M. J.: Factors influencing poor performance in systems analysis and design:
student reflections. In Annual Conference of the Southern African Computer Lecturers' Association,
pp. 251-264. Springer, Cham (2017).
11. Barlow-Jones G, van der Westhuizen D.: Pre-entry Attributes Thought to Influence the Performance of
Students in Computer Programming. In Annual Conference of the Southern African Computer Lectur-
ers' Association, Springer, Cham, pp. 217-226 (2017).
12. Chikumba PA.: Student Performance in Computer Studies in Secondary Schools in Malawi. Interna-
tional Conference on e-Infrastructure and e-Services for Developing Countries, Nov 25 pp. 113-121.
Springer, Berlin, Heidelberg (2010).
13. Akinola, O. S., & Nosiru, K. A.: Factors influencing students’ performance in computer programming:
a fuzzy set operations approach. International Journal of Advances in Engineering & Technology, 7(4),
pp. 1141-1149 (2014).
14. Wilson BC.: A study of factors promoting success in computer science including gender differences.
Computer Science Education. 12(1-2): pp. 141-64 (2002).
15. Rauchas S, Rosman B, Konidaris G, Sanders I. Language performance at high school and success in
first year computer science. In ACM SIGCSE Bulletin Vol. 38, No. 1, pp. 398-402. ACM (2006).
16. Namuye, S., Platz, M., Okanda, P., & Mutanu, L.: Leveraging health through early warning systems
using mobile and service-oriented technology. In 2015 IST-Africa Conference, pp. 1-10, IEEE (2015).
17. Hall M, Frank E, Holmes G, Pfahringer B, Reutemann P, Witten IH.: The WEKA data mining software:
an update. ACM SIGKDD Explorations newsletter. Nov 16;11(1): pp. 10-8 (2009).
18. Team, R. Core.: R: A language and environment for statistical computing. R Development Core Team,
Version 2.6.2, 201 (2013).
19. Kremelberg, D.: Practical statistics: A quick and easy guide to IBM® SPSS® Statistics, STATA, and
other statistical software. SAGE publications, (2010).
20. Kim H, Song HY.: Formulating human mobility model in a form of continuous time Markov chain.
Procedia Computer Science. pp. 389-96 (2012).
21. MacQueen J.: Some methods for classification and analysis of multivariate observations. In Proceedings
of the fifth Berkeley symposium on mathematical statistics and probability Vol. 1, No. 14, pp. 281-297
(1967).
22. Kim T. K.: T-test as a parametric statistic. Korean journal of anesthesiology. 68(6):540 (2015).
23. Sedgwick P.: Pearson's correlation coefficient. BMJ. 345:e4483, (2012).